E-signature was built for a human who clicks

Every product in the category assumes a person in a browser with a mouse. That assumption held for twenty years. It stopped holding about eighteen months ago.

Open any e-signature product and count the decisions that only make sense if a person is sitting there. The drag-and-drop field placer. The prepare-then-review-then-send wizard. The dashboard with a badge on it. The email that says 3 documents awaiting your action.

None of that is bad design. It is correct design for the thing it was designed for: a human, in a browser, with a mouse, doing one document at a time. The category solved that problem completely, and then stopped, because for twenty years there was nothing else to solve for.

What changed

The party assembling the contract increasingly is not a person. It is a model with access to a CRM, a creator database, or an ops spreadsheet, working from an instruction like contract everyone who passed the brief review at the rate we agreed.

Bolting that onto a click-first product produces the worst of both. Browser automation driving a wizard built for a human is fragile, unauditable, and quietly terrifying — a UI change nobody told you about becomes a contract sent to the wrong person.

What an agent needs that a human does not

A person browsing has ambient context an API call does not. They can see the template, remember the fee, notice that the name looks wrong. Strip that away and different things become load-bearing:

• Introspection. A human opens the template and reads it. An agent has to be able to ask what fields exist and which are required.
• A dry run. A human previews by looking. An agent needs a call that renders and validates while writing nothing.
• A structural gate. A human hesitates before sending thirty contracts. An agent does not hesitate, so the irreversible step has to be a separate call that a human authorises.
• Stable correlation. A human recognises a name. An agent needs an identifier it chose, so retries are safe and results are matchable.
• Honest annotations. A client cannot tell which calls are dangerous unless the server says so.

What does not change

The legal machinery underneath is indifferent to who assembled the document. Consent, attribution, integrity, and an audit trail work the same whether a person typed the name or a model extracted it.

Which is the good news: this is not a call to reinvent electronic signature. The evidentiary layer is fine. It is the interface that assumes a mouse.

Where this goes

Signing stays human. Nobody sane wants an agent signing on their behalf, and we deliberately made that impossible — the agent is the sender, never the signer.

But the preparation, the batching, the chasing, the filing, and the triggering of whatever the contract was for — that work was always mechanical, and it was only ever done by a person because there was no one else to do it.

The moment worth protecting is the send. Everything on either side of it is logistics.